R Notebook

Final EDA

Introduction

Credit Card Defaults can occur due to a number of reasons. According to the use case defintion > A default can occur when a borrower is unable to make timely payments, misses payments, or avoids/stops making payments

The question that we are concerend with is > Which priority clients have the highest risk of credit card default?

Libraries

library(tidyverse)

## ── Attaching packages ───────────────────────────────────────────────────────────── tidyverse 1.3.0 ──

## ✓ ggplot2 3.2.1     ✓ purrr   0.3.3
## ✓ tibble  2.1.3     ✓ dplyr   0.8.4
## ✓ tidyr   1.0.0     ✓ stringr 1.4.0
## ✓ readr   1.3.1     ✓ forcats 0.4.0

## ── Conflicts ──────────────────────────────────────────────────────────────── tidyverse_conflicts() ──
## x dplyr::filter() masks stats::filter()
## x dplyr::lag()    masks stats::lag()

library(gmodels)
library(ggridges)
library(reshape2)

## 
## Attaching package: 'reshape2'

## The following object is masked from 'package:tidyr':
## 
##     smiths

library(plotly)

## 
## Attaching package: 'plotly'

## The following object is masked from 'package:ggplot2':
## 
##     last_plot

## The following object is masked from 'package:stats':
## 
##     filter

## The following object is masked from 'package:graphics':
## 
##     layout

Loading Data

train_df <- read.csv("../credit_card_default_train.csv")
train_df$NEXT_MONTH_DEFAULT <- as.factor(train_df$NEXT_MONTH_DEFAULT)
test_df <- read.csv("../credit_card_default_test.csv")
print(nrow(train_df))

## [1] 24000

print(colnames(train_df))

##  [1] "Client_ID"          "Balance_Limit_V1"   "Gender"            
##  [4] "EDUCATION_STATUS"   "MARITAL_STATUS"     "AGE"               
##  [7] "PAY_JULY"           "PAY_AUG"            "PAY_SEP"           
## [10] "PAY_OCT"            "PAY_NOV"            "PAY_DEC"           
## [13] "DUE_AMT_JULY"       "DUE_AMT_AUG"        "DUE_AMT_SEP"       
## [16] "DUE_AMT_OCT"        "DUE_AMT_NOV"        "DUE_AMT_DEC"       
## [19] "PAID_AMT_JULY"      "PAID_AMT_AUG"       "PAID_AMT_SEP"      
## [22] "PAID_AMT_OCT"       "PAID_AMT_NOV"       "PAID_AMT_DEC"      
## [25] "NEXT_MONTH_DEFAULT"

print(str(train_df))

## 'data.frame':    24000 obs. of  25 variables:
##  $ Client_ID         : Factor w/ 24000 levels "A100","A1000",..: 8869 17782 18686 19584 20487 22231 23127 865 1758 2633 ...
##  $ Balance_Limit_V1  : Factor w/ 8 levels " 500K","1.5M",..: 4 4 3 6 4 8 3 3 1 4 ...
##  $ Gender            : Factor w/ 2 levels "F","M": 2 1 1 1 1 1 2 1 2 2 ...
##  $ EDUCATION_STATUS  : Factor w/ 3 levels "Graduate","High School",..: 1 2 2 1 1 1 3 2 3 3 ...
##  $ MARITAL_STATUS    : Factor w/ 2 levels "Other","Single": 1 1 2 2 1 2 2 1 1 2 ...
##  $ AGE               : Factor w/ 4 levels "31-45","46-65",..: 1 3 1 1 1 1 3 3 1 2 ...
##  $ PAY_JULY          : int  -1 0 4 2 2 0 1 2 0 0 ...
##  $ PAY_AUG           : int  -1 -1 3 0 2 0 2 2 0 0 ...
##  $ PAY_SEP           : int  -1 -1 2 0 0 0 2 2 0 2 ...
##  $ PAY_OCT           : int  -1 -1 2 0 0 0 2 0 2 0 ...
##  $ PAY_NOV           : int  -1 -1 -2 0 0 0 2 0 0 0 ...
##  $ PAY_DEC           : int  -1 0 -2 0 0 0 2 2 0 0 ...
##  $ DUE_AMT_JULY      : int  3248 353351 16681 90457 429556 361284 8991 51836 198579 268551 ...
##  $ DUE_AMT_AUG       : int  3389 151818 16082 92848 419466 364802 8515 55828 204634 282726 ...
##  $ DUE_AMT_SEP       : int  6004 26948 15477 95193 429785 366703 11698 54241 218092 274123 ...
##  $ DUE_AMT_OCT       : int  39418 43530 0 97309 435354 353910 11173 55325 212970 221148 ...
##  $ DUE_AMT_NOV       : int  162772 80811 0 100353 445271 356117 12030 59272 213654 222936 ...
##  $ DUE_AMT_DEC       : int  -13982 124590 0 102740 453899 358845 12647 57976 217992 224276 ...
##  $ PAID_AMT_JULY     : int  3437 151818 0 3855 0 16632 0 5521 9240 26565 ...
##  $ PAID_AMT_AUG      : int  6004 46200 0 3890 20790 18480 3696 0 17325 0 ...
##  $ PAID_AMT_SEP      : int  39418 43530 0 3696 16170 12728 0 1984 0 8184 ...
##  $ PAID_AMT_OCT      : int  162772 80811 0 4620 17325 13398 1386 4844 6930 8547 ...
##  $ PAID_AMT_NOV      : int  0 942 0 4049 16401 13860 1155 0 11550 8194 ...
##  $ PAID_AMT_DEC      : int  538165 33666 0 3918 17325 12705 0 2523 11550 7311 ...
##  $ NEXT_MONTH_DEFAULT: Factor w/ 2 levels "0","1": 1 1 2 2 1 1 1 2 1 1 ...
## NULL

print(summary(train_df))

##    Client_ID     Balance_Limit_V1 Gender       EDUCATION_STATUS MARITAL_STATUS
##  A100   :    1   1M     :5951     F: 9540   Graduate   : 8478   Other :13070  
##  A1000  :    1   200K   :5159     M:14460   High School: 3925   Single:10930  
##  A10000 :    1   100K   :3449               Other      :11597                 
##  A10001 :    1   400K   :3065                                                 
##  A10002 :    1    500K  :2790                                                 
##  A10003 :    1   300K   :2411                                                 
##  (Other):23994   (Other):1175                                                 
##            AGE           PAY_JULY           PAY_AUG         PAY_SEP       
##  31-45       :12124   Min.   :-2.00000   Min.   :-2.00   Min.   :-2.0000  
##  46-65       : 4150   1st Qu.:-1.00000   1st Qu.:-1.00   1st Qu.:-1.0000  
##  Less than 30: 7638   Median : 0.00000   Median : 0.00   Median : 0.0000  
##  More than 65:   88   Mean   :-0.01421   Mean   :-0.13   Mean   :-0.1587  
##                       3rd Qu.: 0.00000   3rd Qu.: 0.00   3rd Qu.: 0.0000  
##                       Max.   : 8.00000   Max.   : 8.00   Max.   : 8.0000  
##                                                                           
##     PAY_OCT           PAY_NOV           PAY_DEC         DUE_AMT_JULY    
##  Min.   :-2.0000   Min.   :-2.0000   Min.   :-2.0000   Min.   :-382490  
##  1st Qu.:-1.0000   1st Qu.:-1.0000   1st Qu.:-1.0000   1st Qu.:   8246  
##  Median : 0.0000   Median : 0.0000   Median : 0.0000   Median :  51568  
##  Mean   :-0.2155   Mean   :-0.2612   Mean   :-0.2877   Mean   : 118870  
##  3rd Qu.: 0.0000   3rd Qu.: 0.0000   3rd Qu.: 0.0000   3rd Qu.: 156274  
##  Max.   : 8.0000   Max.   : 8.0000   Max.   : 8.0000   Max.   :2228020  
##                                                                         
##   DUE_AMT_AUG       DUE_AMT_SEP       DUE_AMT_OCT       DUE_AMT_NOV     
##  Min.   :-161185   Min.   :-142079   Min.   :-392700   Min.   :-187882  
##  1st Qu.:   6969   1st Qu.:   6238   1st Qu.:   5429   1st Qu.:   4180  
##  Median :  48717   Median :  46412   Median :  44105   Median :  41863  
##  Mean   : 114073   Mean   : 109244   Mean   : 100357   Mean   :  93777  
##  3rd Qu.: 148905   3rd Qu.: 140162   3rd Qu.: 126975   3rd Qu.: 116926  
##  Max.   :2272881   Max.   :3844046   Max.   :2059564   Max.   :2141765  
##                                                                         
##   DUE_AMT_DEC      PAID_AMT_JULY      PAID_AMT_AUG      PAID_AMT_SEP    
##  Min.   :-784483   Min.   :      0   Min.   :      0   Min.   :      0  
##  1st Qu.:   2913   1st Qu.:   2310   1st Qu.:   1956   1st Qu.:    901  
##  Median :  39409   Median :   4920   Median :   4646   Median :   4197  
##  Mean   :  90341   Mean   :  13306   Mean   :  13867   Mean   :  12093  
##  3rd Qu.: 114435   3rd Qu.:  11605   3rd Qu.:  11550   3rd Qu.:  10626  
##  Max.   :2221444   Max.   :2017905   Max.   :3890638   Max.   :2069852  
##                                                                         
##   PAID_AMT_OCT      PAID_AMT_NOV     PAID_AMT_DEC     NEXT_MONTH_DEFAULT
##  Min.   :      0   Min.   :     0   Min.   :      0   0:18670           
##  1st Qu.:    693   1st Qu.:   610   1st Qu.:    307   1: 5330           
##  Median :   3465   Median :  3465   Median :   3465                     
##  Mean   :  11225   Mean   : 11175   Mean   :  12301                     
##  3rd Qu.:   9360   3rd Qu.:  9412   3rd Qu.:   9252                     
##  Max.   :1434510   Max.   :965557   Max.   :1221218                     
## 

Categorical Variables

First let’s explore the Categorical variables. Since we are interested in what are the factors that affect the risk of credit card default, we shall be exploring considering the two levels of the response variable

Gender

train_df$Gender <- factor(train_df$Gender) # converts to a categorical variable
train_df$NEXT_MONTH_DEFAULT <- factor(train_df$NEXT_MONTH_DEFAULT) # converts to a categorical variable
p1 <- ggplot(data=train_df, aes(x=factor(1), stat="bin", fill=Gender)) + 
  geom_bar(position="fill") # Stacked bar chart
p1 <- p1 + ggtitle("Gender by Next Month Default") + xlab("") + ylab("NEXT_MONTH_DEFAULT") # Adds titles
p1 <- p1 + facet_grid(facets=. ~ NEXT_MONTH_DEFAULT) # Side by side bar chart
p1 <- p1 + coord_polar(theta="y") # side by side pie chart
p1

As it can be clearly seen here, gender does not have a significant difference when compared with defaulting credit cards and not defaulting.

Marital status

train_df$Gender <- factor(train_df$MARITAL_STATUS) # converts to a categorical variable
train_df$NEXT_MONTH_DEFAULT <- factor(train_df$NEXT_MONTH_DEFAULT) # converts to a categorical variable
p2 <- ggplot(data=train_df, aes(x=factor(1), stat="bin", fill=MARITAL_STATUS)) + 
  geom_bar(position="fill") # Stacked bar chart
p2 <- p2 + ggtitle("Marital Status by Next Month Default") + xlab("") + ylab("NEXT_MONTH_DEFAULT") # Adds titles
p2 <- p2 + facet_grid(facets=. ~ NEXT_MONTH_DEFAULT) # Side by side bar chart
p2 <- p2 + coord_polar(theta="y") # side by side pie chart
p2

As it can be clearly seen here, marital status does not have a significant difference when compared with defaulting credit cards and not defaulting. -Hence we have clear reasons for removing the variables gender and marital status_

Education Status

education_response_ct <- CrossTable(train_df$EDUCATION_STATUS,
                               train_df$NEXT_MONTH_DEFAULT ,
                               prop.r= FALSE,
                               prop.c=TRUE,
                               prop.chisq=FALSE, 
                               chisq = TRUE,
                               prop.t=TRUE)

## 
##  
##    Cell Contents
## |-------------------------|
## |                       N |
## |           N / Col Total |
## |         N / Table Total |
## |-------------------------|
## 
##  
## Total Observations in Table:  24000 
## 
##  
##                           | train_df$NEXT_MONTH_DEFAULT 
## train_df$EDUCATION_STATUS |         0 |         1 | Row Total | 
## --------------------------|-----------|-----------|-----------|
##                  Graduate |      6828 |      1650 |      8478 | 
##                           |     0.366 |     0.310 |           | 
##                           |     0.284 |     0.069 |           | 
## --------------------------|-----------|-----------|-----------|
##               High School |      2939 |       986 |      3925 | 
##                           |     0.157 |     0.185 |           | 
##                           |     0.122 |     0.041 |           | 
## --------------------------|-----------|-----------|-----------|
##                     Other |      8903 |      2694 |     11597 | 
##                           |     0.477 |     0.505 |           | 
##                           |     0.371 |     0.112 |           | 
## --------------------------|-----------|-----------|-----------|
##              Column Total |     18670 |      5330 |     24000 | 
##                           |     0.778 |     0.222 |           | 
## --------------------------|-----------|-----------|-----------|
## 
##  
## Statistics for All Table Factors
## 
## 
## Pearson's Chi-squared test 
## ------------------------------------------------------------
## Chi^2 =  63.292     d.f. =  2     p =  1.804334e-14 
## 
## 
## 

Plotting

ggplot(data=as.data.frame(education_response_ct$prop.tbl),
       aes(x=x,y=Freq,fill=y))+
  geom_bar(stat = "identity",position = "fill")+
  scale_y_continuous(labels = scales::percent_format())+
  coord_flip()+theme_minimal()+
  xlab("Education Status")+ylab("Percentages")

Even though it is not evident that there is a relationship between education status and the response variable, the chi square test confirms that there is an association between the two variables ### Balance Limit

blv <- train_df%>% 
  mutate(Balance_Limit_V1=trimws(Balance_Limit_V1)) %>% 
  subset(select=Balance_Limit_V1)
train_df$Balance_Limit_V1 <- factor(as.factor(blv$Balance_Limit_V1),
                          levels=c("100K","200K","300K","400K","500K","1M","1.5M","2.5M"))
balance_response_ct <- CrossTable(train_df$Balance_Limit_V1 ,
                               train_df$NEXT_MONTH_DEFAULT ,
                               prop.r= FALSE,
                               prop.c=TRUE,
                               prop.chisq=FALSE, 
                               chisq = TRUE,
                               prop.t=TRUE)

## 
##  
##    Cell Contents
## |-------------------------|
## |                       N |
## |           N / Col Total |
## |         N / Table Total |
## |-------------------------|
## 
##  
## Total Observations in Table:  24000 
## 
##  
##                           | train_df$NEXT_MONTH_DEFAULT 
## train_df$Balance_Limit_V1 |         0 |         1 | Row Total | 
## --------------------------|-----------|-----------|-----------|
##                      100K |      2207 |      1242 |      3449 | 
##                           |     0.118 |     0.233 |           | 
##                           |     0.092 |     0.052 |           | 
## --------------------------|-----------|-----------|-----------|
##                      200K |      3805 |      1354 |      5159 | 
##                           |     0.204 |     0.254 |           | 
##                           |     0.159 |     0.056 |           | 
## --------------------------|-----------|-----------|-----------|
##                      300K |      1805 |       606 |      2411 | 
##                           |     0.097 |     0.114 |           | 
##                           |     0.075 |     0.025 |           | 
## --------------------------|-----------|-----------|-----------|
##                      400K |      2469 |       596 |      3065 | 
##                           |     0.132 |     0.112 |           | 
##                           |     0.103 |     0.025 |           | 
## --------------------------|-----------|-----------|-----------|
##                      500K |      2289 |       501 |      2790 | 
##                           |     0.123 |     0.094 |           | 
##                           |     0.095 |     0.021 |           | 
## --------------------------|-----------|-----------|-----------|
##                        1M |      5061 |       890 |      5951 | 
##                           |     0.271 |     0.167 |           | 
##                           |     0.211 |     0.037 |           | 
## --------------------------|-----------|-----------|-----------|
##                      1.5M |      1000 |       138 |      1138 | 
##                           |     0.054 |     0.026 |           | 
##                           |     0.042 |     0.006 |           | 
## --------------------------|-----------|-----------|-----------|
##                      2.5M |        34 |         3 |        37 | 
##                           |     0.002 |     0.001 |           | 
##                           |     0.001 |     0.000 |           | 
## --------------------------|-----------|-----------|-----------|
##              Column Total |     18670 |      5330 |     24000 | 
##                           |     0.778 |     0.222 |           | 
## --------------------------|-----------|-----------|-----------|
## 
##  
## Statistics for All Table Factors
## 
## 
## Pearson's Chi-squared test 
## ------------------------------------------------------------
## Chi^2 =  736.0712     d.f. =  7     p =  1.148874e-154 
## 
## 
## 

Plotting

ggplot(data=as.data.frame(balance_response_ct$prop.tbl),
       aes(fill=y,y=Freq,x=x))+
  geom_bar(stat = "identity",position = "fill")+
  scale_y_continuous(labels = scales::percent_format())+
  coord_flip()+theme_minimal()+
  xlab("Balance Limit")+ylab("Percentages")

Here it is clear that a client with a lower balance limit has a higher chance of getting a credit card default. Also from the chi square test it is clear that with a p-value less than alpha=0.05 we can reject H0 and come to the conclusion that there is balance limit and the response variable are not independent

Payment due variable

ggplot(train_df %>%
  gather(PAY_JULY , PAY_AUG , PAY_SEP , PAY_OCT , PAY_NOV , PAY_DEC , key = "PAY_MONTH",value = "NO_MONTHS_PAY"), aes(x = NO_MONTHS_PAY, y = PAY_MONTH, fill = NEXT_MONTH_DEFAULT)) +
  geom_density_ridges(scale=0.9,alpha=0.7) +
  theme_ridges()+
  labs(fill='Credit Default') 

## Picking joint bandwidth of 0.146

Quantitative Variables

Due amounts

ggplot(data=gather(train_df,key,value,c("DUE_AMT_JULY","DUE_AMT_AUG",
                                        "DUE_AMT_SEP","DUE_AMT_OCT",
                                        "DUE_AMT_NOV","DUE_AMT_DEC")) %>% 
         mutate(key = factor(key,levels=c("DUE_AMT_JULY","DUE_AMT_AUG",
                                        "DUE_AMT_SEP","DUE_AMT_OCT",
                                        "DUE_AMT_NOV","DUE_AMT_DEC"))),
       aes(x = key,y = value,fill=NEXT_MONTH_DEFAULT))+
  geom_boxplot()+
  scale_x_discrete(labels=c("JULY","AUG","SEP","OCT","NOV","DEC"))

Paid amounts

ggplot(data=gather(train_df,key,value,c("PAID_AMT_JULY","PAID_AMT_AUG",
                                      "PAID_AMT_SEP","PAID_AMT_OCT",
                                      "PAID_AMT_NOV","PAID_AMT_DEC")) %>%
         mutate(key = factor(key,levels=c("PAID_AMT_JULY","PAID_AMT_AUG",
                                      "PAID_AMT_SEP","PAID_AMT_OCT",
                                      "PAID_AMT_NOV","PAID_AMT_DEC"))),
       aes(x = key,y = value,fill=NEXT_MONTH_DEFAULT))+
  geom_boxplot()+
  scale_x_discrete(labels=c("JULY","AUG","SEP","OCT","NOV","DEC"))

Further visualizations

Paid amount as a ratio of due amount

eps<-0.1
train_df$duepaid_JULY <- train_df$PAID_AMT_JULY/(train_df$DUE_AMT_JULY+eps)
train_df$duepaid_AUG <-  train_df$PAID_AMT_AUG/(train_df$DUE_AMT_AUG+eps)
train_df$duepaid_SEP <-  train_df$PAID_AMT_SEP/(train_df$DUE_AMT_SEP+eps)
train_df$duepaid_OCT <-  train_df$PAID_AMT_OCT/(train_df$DUE_AMT_OCT+eps)
train_df$duepaid_NOV <-  train_df$PAID_AMT_NOV/(train_df$DUE_AMT_NOV+eps)
train_df$duepaid_DEC <-  train_df$PAID_AMT_DEC/(train_df$DUE_AMT_DEC+eps)

train_df$duepaid_JULY[is.nan(train_df$duepaid_JULY)] <- 0
train_df$duepaid_AUG[is.nan(train_df$duepaid_AUG)] <- 0
train_df$duepaid_SEP[is.nan(train_df$duepaid_SEP)] <- 0
train_df$duepaid_OCT[is.nan(train_df$duepaid_OCT)] <- 0
train_df$duepaid_NOV[is.nan(train_df$duepaid_NOV)] <- 0
train_df$duepaid_DEC[is.nan(train_df$duepaid_DEC)] <- 0

new04 <- data.frame(train_df$duepaid_NOV,train_df$duepaid_OCT,
                    train_df$duepaid_SEP,train_df$duepaid_AUG,
                    train_df$duepaid_JULY,train_df$duepaid_DEC,
                    as.factor(train_df$NEXT_MONTH_DEFAULT))
new05 <- new04 %>%
  gather(train_df.duepaid_DEC,train_df.duepaid_NOV,
         train_df.duepaid_OCT,train_df.duepaid_SEP,
         train_df.duepaid_AUG,train_df.duepaid_JULY, key = "month_paid_due",value = "ratio")

new06 <- na.omit(new05)
new06$month_paid_due <- factor(new06$month_paid_due,
                               levels=c("train_df.duepaid_JULY","train_df.duepaid_AUG",
                                        "train_df.duepaid_SEP","train_df.duepaid_OCT",
                                        "train_df.duepaid_NOV","train_df.duepaid_DEC"))
colnames(new06)[1] <- "response"
ggplot(new06, aes(x = month_paid_due , y = ratio)) + 
  geom_boxplot(aes(fill=response))+
  coord_cartesian(ylim=c(0,1))+
  theme_minimal()+
  scale_x_discrete(labels=c("JULY","AUG","SEP","OCT","NOV","DEC"))

### Due amount as a ratio of credit limit

train_df <- train_df %>% mutate(
  Balance_Credit_Limit_Numeric = case_when(
    Balance_Limit_V1 == "100K" ~ 100000,
    Balance_Limit_V1 == "200K" ~ 200000,
    Balance_Limit_V1 == "300K" ~ 300000,
    Balance_Limit_V1 == "400K" ~ 400000,
    Balance_Limit_V1 == "500K" ~ 500000,
    Balance_Limit_V1 == "1M" ~ 1000000,
    Balance_Limit_V1 == "1.5M" ~ 1500000,
    Balance_Limit_V1 == "2M" ~ 2000000,
    Balance_Limit_V1 == "2.5M" ~ 2500000
  )
) %>%
  mutate(
    Due_Credit_Lim_JULY=(DUE_AMT_JULY/Balance_Credit_Limit_Numeric) * 100,
    Due_Credit_Lim_AUG=(DUE_AMT_AUG/Balance_Credit_Limit_Numeric) * 100,
    Due_Credit_Lim_SEP=(DUE_AMT_SEP/Balance_Credit_Limit_Numeric) * 100,
    Due_Credit_Lim_OCT=(DUE_AMT_OCT/Balance_Credit_Limit_Numeric) * 100,
    Due_Credit_Lim_NOV=(DUE_AMT_NOV/Balance_Credit_Limit_Numeric) * 100,
    Due_Credit_Lim_DEC=(DUE_AMT_DEC/Balance_Credit_Limit_Numeric) * 100
  ) %>%
  na.omit()

ggplot(data=gather(train_df,key,value,c("Due_Credit_Lim_JULY","Due_Credit_Lim_AUG",
                                        "Due_Credit_Lim_SEP","Due_Credit_Lim_OCT",
                                        "Due_Credit_Lim_NOV","Due_Credit_Lim_DEC")) %>%
         mutate(key = factor(key,levels=c("Due_Credit_Lim_JULY","Due_Credit_Lim_AUG",
                                        "Due_Credit_Lim_SEP","Due_Credit_Lim_OCT",
                                        "Due_Credit_Lim_NOV","Due_Credit_Lim_DEC"))),
       aes(x=key,y=value))+
  geom_boxplot(aes(fill=NEXT_MONTH_DEFAULT))+
  coord_cartesian(ylim=c(0,200))+
  scale_x_discrete(labels=c("JULY","AUG","SEP","OCT","NOV","DEC"))

### Paid amount as a ratio of credit limit

Balance feature

Here we consider the balance as the Due amount - Paid amount

train_df$balance_july <- train_df$DUE_AMT_JULY - train_df$PAID_AMT_JULY
train_df$balance_aug <- train_df$DUE_AMT_AUG - train_df$PAID_AMT_AUG
train_df$balance_sep <- train_df$DUE_AMT_SEP - train_df$PAID_AMT_SEP
train_df$balance_oct <- train_df$DUE_AMT_OCT - train_df$PAID_AMT_OCT
train_df$balance_nov <- train_df$DUE_AMT_NOV - train_df$PAID_AMT_NOV
train_df$balance_dec <- train_df$DUE_AMT_DEC - train_df$PAID_AMT_DEC

ggplot(data=gather(train_df,key,value,c("balance_july","balance_aug",
                                        "balance_sep","balance_oct",
                                        "balance_nov","balance_dec")) %>%
         mutate(key = factor(key,levels =c("balance_july","balance_aug",
                                        "balance_sep","balance_oct",
                                        "balance_nov","balance_dec"))),
       aes(x = key,y = value))+
  geom_boxplot(aes(fill=NEXT_MONTH_DEFAULT))+
  theme_minimal()+
  scale_x_discrete(labels=c("JULY","AUG","SEP","OCT","NOV","DEC"))

### Balance as a ratio of of credit limit

train_df <- train_df %>% mutate(
    balance_Credit_Lim_JULY=(balance_july/Balance_Credit_Limit_Numeric) * 100,
    balance_Credit_Lim_AUG=(balance_aug/Balance_Credit_Limit_Numeric) * 100,
    balance_Credit_Lim_SEP=(balance_sep/Balance_Credit_Limit_Numeric) * 100,
    balance_Credit_Lim_OCT=(balance_oct/Balance_Credit_Limit_Numeric) * 100,
    balance_Credit_Lim_NOV=(balance_nov/Balance_Credit_Limit_Numeric) * 100,
    balance_Credit_Lim_DEC=(balance_dec/Balance_Credit_Limit_Numeric) * 100
  ) %>%
  na.omit()

ggplot(data=gather(train_df,key,value,c("balance_Credit_Lim_JULY","balance_Credit_Lim_AUG",
                                        "balance_Credit_Lim_SEP","balance_Credit_Lim_OCT",
                                        "balance_Credit_Lim_NOV","balance_Credit_Lim_DEC")) %>%
         mutate(key = factor(key,levels=c("balance_Credit_Lim_JULY","balance_Credit_Lim_AUG",
                                        "balance_Credit_Lim_SEP","balance_Credit_Lim_OCT",
                                        "balance_Credit_Lim_NOV","balance_Credit_Lim_DEC"))),
       aes(x=key,y=value))+
  geom_boxplot(aes(fill=NEXT_MONTH_DEFAULT))+
  coord_cartesian(ylim=c(0,200))+
  scale_x_discrete(labels=c("JULY","AUG","SEP","OCT","NOV","DEC"))

### Pay delay with the Paid value

ggplot(data=train_df%>%
         gather(pay_key,pay_value,c("PAY_JULY","PAY_AUG",
                                        "PAY_SEP","PAY_OCT",
                                        "PAY_NOV","PAY_DEC")) %>%
         gather(paid_key,paid_value,c("PAID_AMT_JULY","PAID_AMT_AUG",
                                      "PAID_AMT_SEP","PAID_AMT_OCT",
                                      "PAID_AMT_NOV","PAID_AMT_DEC")) %>%
         mutate(pay_key = factor(pay_key,levels=c("PAY_JULY","PAY_AUG",
                                        "PAY_SEP","PAY_OCT",
                                        "PAY_NOV","PAY_DEC")),
                pay_value=factor(pay_value,levels=seq(-2,9))),
       aes(y=paid_value,x=pay_value,fill=NEXT_MONTH_DEFAULT))+
  # geom_line(aes(color=NEXT_MONTH_DEFAULT))+
  geom_bar(position="dodge", stat="identity")+
  facet_grid(pay_key ~ .) + theme_minimal()

### Correlation matrix plot

Disregard this as it is too cluttered

train_df_to_cor <- train_df %>% mutate_if(is.factor,as.numeric)
corrmat <- cor(train_df_to_cor,method="spearman")
melted <- melt(corrmat) %>%
  mutate(text = paste0("x: ", Var1, "\n", "y: ", Var2, "\n", "Value: ",round(value,2), "\n"))
p <- ggplot(melted, aes(Var1, Var2, fill= value, text=text)) + 
  geom_tile()
p

ggplotly(p, tooltip="text")